Newsgroups: comp.bugs.sys5
Path: utzoo!geoff
From: geoff@zoo.toronto.edu (Geoffrey Collyer)
Subject: Re: small bug in who(1) of SVR3
Message-ID: <1991Jan14.124459.5096@zoo.toronto.edu>
Organization: U of Toronto Zoology
References: <18896@rpp386.cactus.org> <1991Jan13.004843.18650@zoo.toronto.edu> <18899@rpp386.cactus.org>
Date: Mon, 14 Jan 1991 12:44:59 GMT

> fork() and exec() are neither free nor even cheap - shell scripts are
> just not the right answer.

If memory serves, a PDP-11/70, a Sun 3/180 and a Sun 4/280 (to pick
random examples :-), all with SMD disks, fork and exec at about the
same rate.  Given the increases over time in SMD disk, CPU and memory
speeds, this is inexcusable, and means that relative to hardware
capabilities, fork and exec are getting more expensive over time.  On
reflection, we could account for about 10% of the CPU time taken by the
Sun 3 fork.  (I don't have the numbers at hand; I believe we measured
SunOS 3.5 and 4.0 [with and without shared libraries] where possible.)

If vendors would fix their kernels to stop squandering the available
hardware resources, users could use kernel facilities as they were
intended to be used, rather than twisting and contorting every
application ``because fork and exec are too expensive''.  Fix the
problem in one place.  (One is reminded of the current internet fad for
whapping some combination of authentication and encryption into every
protocol rather than solving the problem in TCP, once, but that's a
flame for another time.)
-- 
Geoff Collyer		utzoo!geoff, zoo.toronto.edu!geoff